ETH Polymer Physics seminar


1999-05-26
10:15 at ML J 34.3

Liquid Brain: A kinetic model of structureless parallel computations

Alexander Gorban

Institute of Computational Modelling SB RAS, 660036, Akademgorodok, Krasnoyarsk, Russia

 The problem of effective programming with fine-grained parallelism is far from being solved. It seems that, despite of numerous efforts, we have not yet understood parallel computations, considering them mainly as a result of a parallelization of the usual algorithms. There are some promising approaches based on models of computing environments constructed from a large number of elementary calculators of the same type (neural networks, cellular automata etc.). If it is possible to implement a problem in such an environment (for example, by methods of neural networks training), further realization with parallel computers can be easily constructed within the framework of the ideas "similar tasks for different elements". There are other perspective ideas and approaches to construction of models of fine-grained parallelism besides neural networks: Parallel Substitution Algorithms (PSA), The chemical computer (SCAM Statistic Cellular Automata Machine), Artificial Immune Systems. A new abstract model of computations, the Kirdin kinetic machine, is presented in this talk. This model is expected to play the same role for parallel computations as the Turing machine and other abstract algorithmic calculators for sequential computations. The Kirdin kinetic machine is based on chemical reactions in liquids or gases. Our optimistic expectations go back to the theorem of M. D. Korzuhin on chemical reactions ability to imitate any dynamic system for finite times, and to the theorem of A.N.Gorban on chemical systems approximating any dynamic systems.


© Mar 2024 mk     719 entries